My experience is that it is best to simulate the effects of finite word length. Calculate various sets of ideal filter coefficients. Then select those sets that have least rounding errors.
Simulate the filter with normal and worst-case input, using the rounded coefficients and with the word length of your target architecture!
This gives a pretty good impression of the reliability.
Best thing is to search in astalavista.com u have got some good stuff toooooooo to get they do have a lot of resources for books and education .u may want to try checking out there